This is a continuation of NAS-107338, which was closed with the comment: "There were two mitigations, one for BETA2 and the other scheduled for RC1."

Since that time, Beta2.1 has been released which appears to include bugfixes after beta2, and therefore I'd guess includes the bugfixes referred to above. But middleware disconnections still persist.

I noticed this on the console within the last hour. A replication ended at 2020-06-09 02.01 (UK time) and another one started at 02:38, 37 minutes later. Between those 2 times the server was utterly idle - no scrubs, no send/recv, no SMB or iSCSI activity whatsoever. Active SSH and console sessions only, virtually idle.

As middelwared log shows, in that brief 37 minute idle span, middleware lost its self-connection twice, at 02.13.10 and again 02.28.10 (and then a third time after send/recv recommenced at 02.43.10). All of these were exactly 15 minutes apart.

After that, further disconnections occurred at 02.48.10, 02.53.10 and 02.58.10, again exactly 5 minutes apart to the second.

In each case the traceback seemed to point to collectd_pyplugins/disktemp.py. and then self.sock.recv. Doubtless the logs have more detail.

I'll have to break the debug file in 2~3 parts to upload, it's got some large files in it that won't fit in the 50M upload limit. I'll upload it shortly or tomorrow as it's late